The neural semiconductor chip first includes: a global register and control logic circuit block, a R/W memory block and a plurality of neurons fed by buses transporting data such as the input vector data, set-up parameters, etc., and signals such as the feed back and control signals. The R/W memory block, typically a RAM, is common to all neurons to avoid circuit duplication, increasing thereby the number of neurons integrated in the chip. The R/W memory stores the prototype components. Each neuron comprises a computation block, a register block, an evaluation block and a daisy chain block to chain the neurons. All these blocks (except the computation block) have a symmetric structure and are designed so that each neuron may operate in a dual manner, i.e. either as a single neuron (single mode) or as two independent neurons (dual mode). Each neuron generates local signals. The neural chip further includes an OR circuit which performs an OR function for all corresponding local signals to generate global signals that are merged in an on-chip common communication bus shared by all neurons of the chip. The R/W memory block, the neurons and the OR circuit form an artificial neural network having high flexibility due to this dual mode feature which allows to mix single and dual neurons in the ANN.
Legal claims defining the scope of protection, as filed with the USPTO.
1. A neural semiconductor chip comprising: a global register and control logic block for storing input data and set-up parameters and for generating control signals for chip operation; a plurality of neurons that are fed in parallel by data and control buses to generate local signals, each neuron including a data bus (DATA-BUS) which transports prototype vector components, an input bus (INPUT-BUS) which transports an actual influence field (AIF) and lower and upper limits thereof (MaxIF/MinIF), a category, a distance and norm/context data, a feed back bus (OR-BUS) which transports feed back signals, computation means, connected to said data bus and input bus and to register means, for performing computation including calculating the distance between an input vector and a stored prototype vector to generate a distance signal, performing comparisons between data stored in said register means or present on said buses, and performing addition/subtraction operations on said data, register means for storing data transported on said buses or generated by said computation means, evaluation means for determining one of a maximum value and a minimum value of data stored in at least one of said register means and R/W memory means or present on said buses, daisy chain means, connected to corresponding daisy chain means of two adjacent neurons when each said neuron is part of an artificial neural network (ANN), for differentiating between different possible states of each said neuron (free, first free and engaged), wherein said register means, said evaluation means and said daisy chain means have a substantially symmetric structure, R/W memory means, common to all neurons, cut into slices for storing the prototype components of a neuron, adapted to store prototype vector components, and OR means for performing an OR function of all local signals outputted by said neurons to generate a global signal that is re-injected as a feed back signal into each neuron via the feed back bus; and logic control means for causing a neuron to operate either as a single neuron (single mode) or as two independent neurons (dual mode) characterized as even and odd neurons.
2. The neural semiconductor chip of claim 1 wherein said R/W memory is a RAM memory and every slice includes P M bits where P is the number of addresses and M is the number of bits to code a prototype component.
3. The neural semiconductor chip of claim 2 wherein P represents the maximum number of prototype components of a single neuron (single mode).
4. The neural semiconductor chip of claim 2 wherein P/2 represents the maximum number of prototype components of either the even or the odd neuron (dual mode).
5. The neural semiconductor chip of claim 4 wherein in the RAM the lower half addresses represent the even neuron and the lower half addresses represent the odd neuron in the dual mode.
6. The neural semiconductor chip of claim 2 further comprising: means for writing only in a slice and not in others slices of the RAN memory.
7. The neural semiconductor chip of claim 2 further comprising: means for writing bits only in a specific part of a slice.
8. The neural semiconductor chip of claim 2 wherein said writing means allows to write either in the MSBs or in the LSBs of a determined address of the RAM memory slice to increase the number of prototype components stored at a determined address but with a lower precision.
9. The neural semiconductor chip of claim 1 further comprising: masking means for feeding masked data to a neuron.
10. The neural semiconductor chip of claim 9 wherein said masking means comprises: register means to store the mask data; and, AND gate means connected between the R/W memory and the improved neuron receiving the mask data on a first input and the data stored in the R/W memory on a second input.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
December 22, 1999
February 18, 2003
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.